WO2021019551A1

WO2021019551A1 - System and method for real-time self-optimization of manufacturing operations

Info

Publication number: WO2021019551A1
Application number: PCT/IN2020/050522
Authority: WO
Inventors: Vishnu Swaroopji Masampally; Aditya Pareek; Chetan Bharat JADHAV; Venkataramana Runkana
Original assignee: Tata Consultancy Services Limited
Priority date: 2019-07-26
Filing date: 2020-06-12
Publication date: 2021-02-04

Abstract

This disclosure relates generally to method and system for real-time self-optimization of manufacturing operations and systems. Generally the behavior of a manufacturing plant changes with time due to the regime changes or aging of equipment. Subsequently, the optimization configuration of the plant needs to be changed such that it suits the current plant behavior. The disclosed system identifies the change in plant behavior by monitoring disturbance variables and KPIs of the plant. Once the change in the plant exceeds a critical limit quantified in terms of changes in disturbance and process variables along with key performance indicators, the current optimization problem is deemed no longer valid, and the system triggers self-optimization of the plant. The system configures a new optimization problem considering the earlier knowledge of problem formulation and the current plant conditions.

Description

SYSTEM AND METHOD FOR REAL-TIME SELF- OPTIMIZATION OF

MANUFACTURING OPERATIONS

CROSS-REFERENCE TO RELATED APPLICATIONS AND PRIORITY

[001] The present application claims priority from Indian patent application no. 201921030370, filed on luly 26, 2019. The entire contents of the aforementioned application are incorporated herein by reference.

TECHNICAL FIELD

[002] The disclosure herein generally relates to the field of modeling and self optimization of manufacturing operations, and, more particularly, to system and method for self-optimization of manufacturing operations in real-time.

BACKGROUND

[003] A manufacturing plant and the accompanying processes, in general, consist of various units. Feed is processed in each of the units where it is transformed either chemically or physically or both. Each such unit or a group of units can have recycle streams or purge streams or addition of utilities (exogenous inputs to facilitate physical/chemical transformation) or combination of these streams. Such plants, in general, are designed such that, the final product obtained after processing the feed and the intermediate products obtained through multiple units needs to meet the required performance criteria such as productivity, quality, quantity, etc. It is also expected that the product loss in any of the purge stream is minimum. Moreover, each unit or equipment is operated in a safe (low environment impact) and efficient (low utility consumption) condition.

[004] However, demand of a product, its specifications, quality parameters, emission requirements are rarely immutable and need to be updated dynamically based on business requirements and government regulations. In addition, quality of the raw material to be processed in the plant too varies. Equipment installed in a manufacturing plant, that processes the raw material and other intermediates, ages and deviations are thus observed from its expected behavior. Therefore, the operating space of the plant widens and deviates significantly from what it was initially designed for. The dynamic changes in the required specifications of the key plant / process outputs, equipment condition, along with the uncontrollable inputs, referred to as disturbance variables, such as raw materials and environmental conditions, result in plant being operated sub-optimally. To improve the plant performance decision variables, which are mostly exogenous inputs and processing decisions that influence the performance of equipment and therefore a plant, need to be updated. This is generally done manually by plant operators or process engineers assisted by process knowledge gathered with experience. The interconnected and complex nature of plant operations makes the decision making highly complicated, improbable and error prone, if done manually, and not in- sync with the process requirements. At present, an autonomous system that prescribes the decision variables and processing inputs for manufacturing plant / processes in such dynamically changing environment does not exist

SUMMARY

[005] Embodiments of the present disclosure present technological improvements as solutions to one or more of the above-mentioned technical problems recognized by the inventors in conventional systems. For example, in one embodiment, a processor-implemented method for real-time self-optimization of manufacturing processes is provided. The method includes determining a plurality of optimal set points of decision variables for current state of disturbance variables associated with a plurality of manufacturing processes for optimization of key process outputs of the plurality of manufacturing processes, the optimization performed by using one or more optimization models selected from an optimization model database. The method includes self-monitoring components that decides when to trigger optimization. Further, the method includes obtaining current state of the decision variables and one or more disturbance variables, and actual outputs of the plurality of manufacturing processes from a plurality of source databases and soft sensors. Also, the method includes selecting at least one soft sensor from amongst a plurality of soft sensors stored in a soft sensor repository based on the actual outputs of the plurality of manufacturing processes, each of the plurality of soft sensors associated with a corresponding prediction model and capable of predicting process outputs of a manufacturing process from amongst the plurality of manufacturing processes.

[006] Furthermore, the method includes predicting expected process output of a manufacturing process, using at least one soft sensor corresponding to the process output or outputs that are being optimized. The actual process outputs are compared with the expected process outputs to calculate deviation of the expected outputs from the actual outputs. The expected process outputs are compared with desired process outputs to calculate deviation of the expected outputs from the desired process outputs, the desired outputs associated with the plurality of manufacturing processes stored in a knowledge base. Real-time adaptive tuning of the one or more optimization models is enabled on determination of the deviation of the expected process outputs from the desired process outputs being more than a first predefined threshold. Either a notification to update the model is generated or real-time adaptive tuning of one or more of the plurality of soft sensors and the corresponding prediction models is enabled on determination of the deviation of expected process outputs compared to actual process outputs being more than a second predefined threshold.

[007] In another aspect, a system for real-time self-optimization of manufacturing processes is provided. The system includes one or more memories; and one or more hardware processors, where the one or more memories are coupled to the one or more hardware processors, and wherein the one or more hardware processors are capable of executing programmed instructions stored in the one or more memories to determine a plurality of optimal set points of decision variables for current state of disturbance variables associated with a plurality of manufacturing processes for optimization of key process outputs of the plurality of manufacturing processes, the optimization performed by using one or more optimization models selected from an optimization model database. Further, the one or more hardware processors are capable of executing programmed instmctions to obtain current state of the decision variables and one or more disturbance variables, and actual outputs of the plurality of manufacturing processes from a plurality of source databases. Also, the one or more hardware processors are capable of executing programmed instructions to select at least one soft sensor from amongst a plurality of soft sensors stored in a soft sensor repository based on the actual outputs of the plurality of manufacturing processes, each of the plurality of soft sensors associated with a corresponding prediction model and capable of predicting output of a manufacturing process. Furthermore, the one or more hardware processors are capable of executing programmed instmctions to predict expected process outputs of manufacturing process, using at least one soft sensor, expected output of a manufacturing process associated with the at least one soft sensor. The actual process outputs are compared with the expected outputs to calculate the deviation of the expected outputs compared to the actual process outputs. The expected process outputs are compared with desired outputs to calculate the deviation of the expected process outputs from the desired outputs, the desired outputs associated with the plurality of manufacturing processes stored in a knowledge base. Real-time adaptive tuning of the one or more optimization models is enabled on determination of the deviation of the expected process outputs compared to the desired outputs being more than a first predefined threshold. A real-time adaptive tuning of one or more of the plurality of soft sensors and the corresponding prediction models is enabled on determination of the deviation of expected outputs compared to actual outputs being more than a second predefined threshold.

[008] In yet another aspect, a non-transitory computer readable medium for real-time self-optimization of manufacturing processes are provided. Said one or more non-transitory machine readable information storage mediums comprises one or more instructions which when executed by one or more hardware processors causes determining a plurality of optimal set points of decision variables for a current state of disturbance variables associated with a plurality of manufacturing processes for optimization of output of the plurality of manufacturing processes, the optimization performed by using one or more optimization models selected from an optimization model database. Further the method includes obtaining a current state of the decision variables and one or more disturbance variables, and actual outputs of the plurality of manufacturing processes from a plurality of source databases. Also, the method include selecting at least one soft sensor from amongst a plurality of soft sensors stored in a soft sensor repository based on the actual outputs of the plurality of manufacturing processes, each of the plurality of soft sensors associated with a corresponding prediction model and capable of predicting output of a manufacturing process from amongst the plurality of manufacturing processes. Furthermore, the method includes predicting, using the at least one soft sensor, expected output of a manufacturing process associated with the at least one soft sensor. The actual outputs are compared with the expected outputs to identify a deviation of the expected outputs compared to the actual outputs. The expected outputs is compared with desired outputs to identify a deviation of the expected outputs compared to the desired outputs, the desired outputs associated with the plurality of manufacturing processes stored in a knowledge base. A real-time adaptive tuning of the one or more optimization models is enabled on determination of the deviation of the expected outputs compared to the desired outputs being more than a first predefined threshold. A real-time adaptive tuning of one or more of the plurality of soft sensors and the corresponding prediction models is enabled on determination of the deviation of expected outputs compared to actual outputs being more than a second predefined threshold.

BRIEF DESCRIPTION OF THE DRAWINGS

[009] The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed principles: [010] The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed principles:

[Oil] FIG. 1A illustrates an exemplary networking environment implementing a system for real-time self-optimization of manufacturing operations according to some embodiments of the present disclosure.

[012] FIG. IB represents a manufacturing plant for real-time self-optimization in accordance with an example embodiment of present disclosure.

[013] FIG. 2 is flow diagram illustrating a method for creating an optimization model according to some embodiments of the present disclosure.

[014] FIG. 3 is flow diagram illustrating a method for extraction of lower and upper bounds of decision variables using the variables information according to some embodiments of the present disclosure.

[015] FIG. 4 is flow diagram illustrating a method for creation of objective and constraint functions, key constituents of an optimization model, using the variables’ information and soft sensors according to some embodiments of the present disclosure.

[016] FIG. 5 is flow diagram illustrating a method for classification of optimization problem according to some embodiments of the present disclosure.

[017] FIG. 6 is flow diagram illustrating a method for real time optimization according to some embodiments of the present disclosure.

[018] FIG. 7 is flow diagram illustrating a method for estimating the current state of disturbance variables for optimization according to some embodiments of the present disclosure.

[019] FIG. 8A is flow diagram illustrating a method for enabling self-monitoring in the plant according to some embodiments of the present disclosure.

[020] FIG. 8B is flow diagram illustrating a method for enabling self-monitoring in the plant according to another embodiment of the present disclosure.

[021] FIG. 9 is flow diagram illustrating a method for self-optimization according to some embodiments of the present disclosure.

[022] FIGS. 10A and 10B illustrate a flow diagram for a method for real-time self optimization of manufacturing processes according to some embodiments of the present disclosure.

[023] FIG. 11 is a block diagram of a system for real-time self-optimization of manufacturing plant according to some embodiments of the present disclosure. DETAILED DESCRIPTION OF EMBODIMENTS

[024] Optimization of manufacturing plant operations and/or processes refers adjusting processes and/or systems associated with the plant in such a manner to optimize a set of key process outputs without violating certain constraints that are again expressed in terms of key process outputs. Typically, optimization is performed with an objective of minimizing cost of operation and maximizing throughput and/or efficiency.

[025] Rapid and extensive digitization of the manufacturing plants has made available information/data pertaining to said plant operations for real time consumption and manipulation. Soft sensors, predictive models developed using either physics or Artificial Intelligence (AI) techniques, are being deployed to monitor key process outputs and equipment condition. In the present disclosure, a self-optimization system that utilizes process knowledge that is already captured in the soft sensors and the various plant databases, in order to perform model-based optimization of key process outputs and determine set points of the decision variables under dynamically changing environment of the manufacturing plant is presented. A real-time self-optimization system is presented in FIG. 1A.

[026] FIG. 1 A illustrates a representation of a real-time self-optimization system 100 for a plant for example, a manufacturing plant 120 in accordance with an example embodiment of present disclosure. The real-time self-optimization system 100 includes a real-time optimization module 102, a simulation module 104, a self-monitoring module 106, a retuning module 108, a knowledge base 110, an optimization model repository 110a, a prediction model repository 110b and a soft sensor repository 110c. A plant database 122 interacts with plant 150 via various systems like distributed control system (DCS), laboratory information management system (LIMS), and so on and maintains real-time data, material quality data, equipment design and maintenance data, and so on. The plant database 122 interacts with plant 150 also facilitates in providing the set-points and suggestions to the plant 120. The real-time self-optimization system 100 interacts with the plant database 122 regularly in order to provide the set points obtained from the real-time optimization module 102 such that it can be implemented as set-points by the lower level controllers. On the other hand, real-time self optimization system 100 also interacts with the plant 120 in order to provide the expected values of the key performance indicators (KPIs) estimated using the simulation module 104. The realized values of KPIs from the plant 120 are compared to the expected values of KPIs from the simulation module 104 at the self -monitoring module 106 for monitoring the performance of soft- sensors. The accuracy of the optimization models is also monitored at self- monitoring module 106 by comparing the realized KPIs to the best possible KPIs stored in the knowledge base for given disturbance variables. Retuning module 106 is triggered whenever the error during comparison exceeds a predetermined threshold value. The retuning module 108 retunes or reconfigures either soft- sensors or optimization models and updates them in the knowledge base 110 for further use. All the information required for the real-time self optimization system is stored in the knowledge base 110. This knowledge base utilizes the existing optimization model repository 110a, prediction model repository 110b and soft-sensor repository 110c to perform real-time self-optimization. A typical manufacturing plant is described further with reference to FIG. IB.

[027] FIG. IB represents a typical manufacturing plant 170 that would be benefitted by implementation of real-time self-optimization in accordance with an example embodiment of present disclosure. The manufacturing plant 170 may hereinafter be referred to as a plant 170. It should be noted that plant can consist of a single or multiple units, such units are equipment with each one serving its own purpose, and can have these units connected in any arbitrary manner as required to manufacture product of interest. Furthermore, the product manufactured is not limited to a drug product (both small and large molecules), cement, specialty chemical, concentrated mineral, refined metal, refined oil, petroleum products, and so on. The manufactured product can also refer to a utility such as electrical energy, wind energy, solar energy, chemical energy etc. In the given text, such a manufacturing plant is shown to include various units such as units numbered 1-9 for the purpose of processing the mineral ore feed to produce concentrated mineral. The external process inputs, representing some of the decision variables, are shown in form of dotted arrows.

[028] In an example scenario, the plant 170 takes a low concentration mineral ore as feed and enriches it to a high concentration product. The enriched product (or mineral ore) is collected as concentrate from unit 8 and 9. The unwanted minerals are separated from the feed and are removed via tails from unit 9.

[029] The feed is a mineral ore that consists of low concentrations of required mineral. The mineral processing plant 170 helps enriching the mineral content in the feed using froth flotation technique. In order to process the feed for froth flotation, the size of the feed is reduced using various crushing and grinding techniques. The feed is initially fed to primary crushing at unit 1, a SAG mill in general. The product from unit 1 is screened at unit 2 using vibratory screens. The mineral ore lumps that are less than a required size is fed to unit 4 for further processing. The lumps that are above the required size are further crushed at unit 3, a pebble crusher in general. These crushed lumps are again fed to unit 1. At unit 4, the mineral ore is mixed with water, in general in a sump. The product from unit 4 is pumped to unit 6 using slurry pumps at unit 5. At unit 6, the slurry is separated to fines and coarse using hydro cyclones. The fines are fed to unit 8 for forth flotation. The coarse are fed to grinding at unit 7, a ball mill in general. The fines from unit 7 are dumped into the sump or unit 4.

[030] The fines obtained from unit 6 are further processed for mineral extraction using techniques like froth flotation at unit 8 and unit 9. The tails from unit 8 are passed to unit 9 for further extraction of the required mineral. The concentrate from both unit 8 and 9 are considered to be the final product. There can be few more units for froth flotation based on the required grade of the final product. On the other hand, there can be other units for similar separation techniques or different separation techniques like gravity separation.

[031] All the units from 1 to 9 are collectively considered a mineral processing plant, for example the plant 120 (FIG. 1A), and this plant interacts with the proposed real-time self optimization system, for example the real-time self-optimization system 100 (FIG. 1A). In the case of a mineral processing plant, the important KPIs, in general, are particle size distribution (PSD), grade and recovery of the final product. The soft-sensors that are built for these KPIs help in monitoring the PSD, grade and recovery of the product in real-time using the simulation module (for example the simulation module 104) of the real-time self-optimization system 100 (FIG. 1A). The real-time optimization module 102 (FIG. 1A) suggests the optimal set points for a given feed quality. These set points could be for decision variables like feed flowrate at each unit, amount of water to be added to each stream, amount of additives and frother to be added at unit 8 and 9, etc. On the other hand, the disturbance variables could be the feed grade, feed particle size distribution, etc. However, if the accuracy of these soft-sensors is too less or the set-points prescribed by the real-time optimization module 102 (FIG. 1 A) are not providing best possible grade and recovery, the soft-sensors and / or the optimization models are retuned or reconfigured. The new information created or information modified in the process is updated to the knowledge base, for example the knowledge base 110 (FIG. 1 A).

[032] Various embodiments disclosed herein overcome the above-mentioned technical challenges and the various other limitations associated with the optimization of operations and processes in a manufacturing plant. For example, the embodiments provide method and system for self-monitoring of plant operations in terms of identification of any changes in the critical external factors, referred to as disturbance variables, that can influence plant behavior, and in the key process outputs of the plant that determine the performance of the plant. In case, the disclosed system identifies that the change in the plant conditions exceeds a critical limit quantified in terms of changes in disturbance and key process outputs, the disclosed system triggers self-optimization of the plant. For optimization, the disclosed system allows and assists the configuration user or operator to define an optimization model by creating objective and constraint functions using the available soft-sensors of key process outputs. For self-optimization, the disclosed system automatically configures and evaluates the defined optimization model for the new current operating conditions of the plant. In an embodiment, the disclosed system performs real-time self-optimization of the key process outputs of the plant and updates set points for the decision variables. Hereinafter, for the brevity of description, the system for real-time self-optimization may be referred to as RTO.

[033] Exemplary embodiments are described with reference to the accompanying drawings. In the figures, the left- most digit(s) of a reference number identifies the figure in which the reference number first appears. Wherever convenient, the same reference numbers are used throughout the drawings to refer to the same or like parts. While examples and features of disclosed principles are described herein, modifications, adaptations, and other implementations are possible without departing from the spirit and scope of the disclosed embodiments. It is intended that the following detailed description be considered as exemplary only, with the true scope and spirit being indicated by the following claims.

[034] Referring now to the drawings, and more particularly to FIG. 1 through 11, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments and these embodiments are described in the context of the following exemplary system and/or method.

[035] The written description describes the subject matter herein to enable any person skilled in the art to make and use the embodiments. The scope of the subject matter embodiments is defined by the claims and may include other modifications that occur to those skilled in the art. Such other modifications are intended to be within the scope of the claims if they have similar elements that do not differ from the literal language of the claims or if they include equivalent elements with insubstantial differences from the literal language of the claims.

[036] As described herein, the system 100 performs self- monitoring, real-time optimization, and self-optimization of operations and systems associated with a plant, for example, a manufacturing plant. Herein, it will be noted that the‘plant’ refers to manufacturing plant or any other facility that may involve interaction of a plurality of operations that have to managed together so as to operate the plant in a most optimized manner.

[037] The system 100 optimizes (either maximizes or minimizes) one or more key process outputs while maintaining constraints expressed in terms of a sub-set of process outputs within defined limits. The system 100 further prescribes set-points of the decision variables to the lower level controllers to be implemented in the plant. However, it is expected that the plant behavior changes with time due to various factors, including but not limited to, the regime changes and/or aging of equipment. Herein, the term‘regime’ refers to operational information including, but not limited to performance and design of process equipment, raw material and environmental conditions, which are parameterized and may be stored / updated in plant database.

[038] Subsequently, the optimization configuration may have to be changed such that it suits current (or changed or altered) plant behavior. In an embodiment, the system 100 identifies a change in plant behavior by monitoring disturbance variables and subset of process outputs of the plant. Once the change in the plant exceeds a critical limit quantified in terms of changes in disturbance and process variables along with the key performance indicators, the current optimization problem is deemed no longer valid, thereby triggering self-optimization of the system.

[039] During self-optimization, the system 100 measures state of disturbance variables and subset of process outputs of the plant, and self-triggers real-time optimization thereof based on change in the state of disturbance variable, and / or difference of actual/predicted values of the KPI and desired value of KPI. The system 100 extracts the information of disturbance variables, which define the state of the disturbance variable, in terms of signature such as present value, or mean and standard deviation, median and inter quantile range, and/or exponential moving average based on process timescale and computational time of the optimization problem. The system 100 executes optimization model that consists of the optimization problem along with appropriate optimization- solver and corresponding solver parameters. The solver here represents the algorithm used to solve the optimization problem defined in the optimization model. The solver parameters depend on the optimization algorithm used. For example, if optimization algorithm is particle swarm optimization (PSO), the parameters can be number of particles, acceleration coefficient, maximum velocity, and number of neighbors for each particle. Each optimization solver can have its unique set of solver parameters.

[040] For self-optimization, the system 100 configures a new optimization problem by extracting the required information from the soft sensors and knowledge base. Further, the category of new optimization problem is obtained by classifying it in order to select a solver. Said knowledge base captures the knowledge that is being created by the system 100. The knowledge base 110 includes a plurality of soft sensors and corresponding annotations thereof, datasets used to build models, names of available variables or parameters, minimum and maximum of variables or parameters for different regimes of data, variable type (decision variable or disturbance variable), variable data type (discrete or continuous), desired process outputs for different regimes of plant operation and the corresponding optimal set-points of decision variables. The above mentioned operating regimes define various operating regions or conditions that can be classified such that a proper distinction is brought out based on the either plant performance or anything as such. These regimes can also characterized by different steady states of the manufacturing plant/equipment/process.

[041] The knowledge base 110 is updated in real-time whenever any component of knowledge base is changed. The knowledge base is updated by saving and/or updating the information in real-time. For example, the knowledge base is updated when one of the plurality of the soft-sensors is created or updated or deleted, one of the plurality of the prediction model is created or updated or deleted, one of the plurality of the dataset used to create a prediction model is created or updated or deleted, one of the plurality of the information regarding variables (like whether a variable is a decision variable or a disturbance variable) is created or updated or deleted, one of the plurality of the statistical metrics of data for different regimes is created or updated or deleted, one of the plurality of the variable type (for example, decision variable or disturbance variable) is created or updated or deleted, one of the plurality of the variable datatype (for example, discrete or continuous) is created or updated or deleted, one of the plurality of the desired process outputs for different regimes of plant operation is created or updated or deleted, one of the plurality of the optimal set-points for different regimes of plant operation is created or updated or deleted, one of the plurality of the optimization model is created or updated or deleted.

[042] In an embodiment, the system 100 creates objective functions and constraint functions from a plurality of soft-sensors defined and accessible to the system 100, and extracts the attributes of the variables from said soft-sensors to thereby define optimization problem such as lower and upper bounds, and type of variable either continuous or discrete. In an embodiment, the system 100 defines multi-model optimization problem by combining outputs/predictions of multiple soft sensors. Herein ‘soft-sensors’ refers to computer algorithms or programs that are capable of estimating quality indexes (or process outputs) of a plant that may otherwise be difficult or expensive to measure in real-time; or that are may otherwise be necessary to be utilized in optimization. Soft sensors includes prediction models that provide scenarios in which estimations can drive decision-making, facilitate in monitoring the plant and improve the reliability of current systems in the plant, often working hand-in- hand with their hard-sensor counterparts, creating comprehensive monitoring networks. In an embodiment, the soft sensor of a process output in the soft sensor repository includes annotations needed by optimization model, and a prediction model extracted from a prediction model repository. The prediction model extracted from the prediction model repository includes at least one of physics based model, data-driven model, and hybrid model. The hybrid model is a combination of physics based model and data-driven model.

[043] The soft sensor uses a combination of historical process data recorded from online sensors along with any laboratory measurements to predict KPIs, thereby replacing manual testing. The soft sensors make use of (1) secondary variables that may be either measured or estimated in real-time and a (2) mathematical model that may correlate these parameters and the variables to be monitored. Typically, a soft sensor is composed of a process model, names of variables and parameters used by the model for prediction, and an update technique. A mathematical model can be classified in terms of the process knowledge within it, going from the physics based or first principle based equations to pure multivariate statistical methods. First principles models such as mechanistic and phenomenological models are very knowledge-intensive and are referred as“white box”, while multivariate models are data- intensive, and usually named as“black box”. Flerein, the disclosed system possesses the ability to handle both physics based and machine learning models.

[044] An important contribution of the disclosed embodiments is self-identification capability of the system 100 that enables the system 100 to identify type of optimization problem based on the nature of the objective function, constraint functions and the variables involved. The system 100 identifies appropriate solver(s) based on the nature of identified objective function and constraint functions.

[045] In an embodiment, the system 100 may be implemented in a network environment such as an IoT based environment comprising various hardware and software elements collectively configured to perform real-time self-optimization of manufacturing plant operations and systems, according to an exemplary embodiment of the disclosure. The IoT based platform backend may include a cloud server connected to the knowledge base, for example, the knowledgebase 110. The system 100 further includes various IoT based devices implemented on different smart devices such as smart phone, a telematics device, and so on enabling real-time analytics of sensor data. The system further includes various heterogeneous sensor devices placed in the vicinity of smart computing environment connected with various IoT based devices. Alternatively, said sensor devices may be embodied in the IoT based devices. Thus, the sensor devices along with the IoT based devices may collectively form an intelligent smart environment or a digital twin of the plant according to this exemplary embodiment. A‘digital twin’ refers to a digital replica of physical assets, process, people, places, systems and devices. The digital representation provides both the elements and the dynamics of how IoT devices operate. In various embodiments, the digital twin of the plant provides current operating conditions of various elements, systems and subsystems of the plant.

[046] A flow diagram explaining steps involved in functioning of the disclosed system 100 is described further in detail with reference to FIG. 2.

[047] Referring now to FIG. 2, a flow diagram of a method 200 for real-time self optimization of a plant is described, in accordance with an example embodiment. The method 200 may be described in the general context of computer executable instructions. Generally, computer executable instructions can include routines, programs, objects, components, data structures, procedures, modules, functions, etc., that perform particular functions or implement particular abstract data types. The method 200 may also be practiced in a distributed computing environment where functions are performed by remote processing devices that are linked through a communication network. The order in which the method 200 is described is not intended to be construed as a limitation, and any number of the described method blocks can be combined in any order to implement the method 200, or an alternative method. Furthermore, the method 200 can be implemented in any suitable hardware, software, firmware, or combination thereof. In an embodiment, the method 200 depicted in the flow chart may be executed by a system, for example, the system 100 of FIG. 1A. In an example embodiment, the system 100 may be embodied in an exemplary computer system.

[048] The entire process of RTO, self-monitoring and retuning requires soft sensors, which can predict the KPIs, and optimization models that can prescribe set points of manipulated variables. For configuration of soft-sensors, the system 100 (FIG. 1 A) defines and saves soft sensors in such a way that it can serve both monitoring and optimizing purposes. The inputs required to define a soft-sensor for a process output, may include but are not limited to: a) a prediction model that can either be a first principle based or data driven or a hybrid model that is capable to predict process output, b) locations of independent soft-sensors used in building the model, and c) miscellaneous inputs required by the prediction model of the soft- sensor. Among the inputs required to define the soft sensor, the prediction model is used for predicting desired variable and rest all inputs serve as inputs to the prediction model for prediction of process output. Using the above-mentioned inputs, the system 100 (FIG. 1A) saves the soft-sensor of a KPI in the soft-sensor repository. As previously described, a soft- sensor also includes annotations along with a prediction model. These annotations are required for creating an optimization model from the prediction models assigned to selected soft-sensor. The prediction models are stored in the prediction model repository 110a (FIG. 1A) and are utilized by the respective soft-sensors whenever a soft-sensor is triggered for prediction. The prediction model extracted from the prediction model repository includes at least one of physics based model, data-driven model, or a hybrid model, with a hybrid model being a combination of both physics based and data-driven models. In an embodiment, the prediction models extracted from the prediction model repository predict process outputs of the manufacturing processes.

[049] For real-time prediction, the soft-sensor requires values of physical sensors in real-time. Therefore, real-time sensor data along with the corresponding soft- sensor of the process output is an input to the simulation module. The soft-sensor may also require predictions from other soft sensors which are used in defining a dependent soft-sensor. These soft-sensors that are used for building the dependent soft-sensor are known as independent soft- sensors and their locations along with their names are passed as an input for simulation module. On the other hand, the soft-sensor may also need other miscellaneous inputs like a configuration file with model parameters, or an executable file that is utilized inside the soft sensor or some standard tables, which helps to calculate outputs such as calculating cost of operations. Said miscellaneous inputs are also passed as an input to simulation module along with the corresponding soft-sensor. The disclosed system, for instance the system 100 tags the aforementioned inputs to the soft sensor and the tagged soft sensor to the soft sensor repository.

[050] Although the above information regarding predicting key process outputs may be sufficient for monitoring of the plant, certain other information may be required for optimization of the plant operations and systems. Therefore, along with the above-mentioned inputs, prediction model annotations may be tagged to the soft-sensor while saving it. Said prediction model annotations may be tagged and saved along with the soft-sensor in the soft- sensor repository. Examples of annotations of prediction model may include, but are not limited to, name of response or output variable, names of features/variables, and model parameters used to build model, data used to train a data-driven model or tune the parameters of the first principle based model, minimum and maximum of decision variables of plant variables (inferred from data or user specified), representative state of disturbance variables (inferred from data), and identification of data type of variables as either discrete or continuous (inferred from data).

[051] In an embodiment, optimization of the KPIs is performed in the plant and set points are prescribed for the decision variables in the plant. However, since plant behavior changes with time, the optimization models and/or soft-sensors for optimization and the optimization configuration itself needs to be changed such that it suits the current plant behavior.

[052] It will be understood that prior to executing real-time self-optimization, optimization model needs to be defined or configured. The configuration or definition of an optimization model refers to creation or update of objective and/or constraint functions along with identification of suitable optimization solver and its parameters. The optimization model has to be configured in two different instances: a) during the initial definition of optimization problem to optimize process outputs within feasible space using the available soft-sensors b) when the retuning of optimization model is triggered as instructed by self-monitoring module. In either case, the optimization model has to be evaluated as described with reference to FIG. 2.

[053] Referring now to FIG. 2, a process flow describing a method 200 for defining and creating an optimization model for RTO is disclosed, in accordance with an example embodiment. At 202, the method includes identifying and extracting, from the soft sensor repository 204, the prediction models and the annotations for soft sensors corresponding to the process outputs that are selected as objective and constraint functions. In an embodiment, those soft-sensors are selected such that corresponding process outputs of the plant can be minimized or maximized as desired. In an embodiment, the selection of the soft-sensors may be done manually. For instance, the soft-sensor that estimates total cost of operations may be selected so that total cost of manufacturing process can be minimized. Additionally or alternatively, the soft sensors that estimate final product quality indices or attributes may be selected so that the quality of product (for instance the effluent stream of FIG. IB) can he maximized. The selection of soft-sensors is dependent on the KPIs that are being optimized and/or constrained. The soft- sensors over which constraints have to be defined are selected along with their limits (namely, lower operating point (LOP) and higher operating point (HOP)). Using annotations of all the soft-sensors that are selected as either objective or constraint function, a unique super-set of variables that are needed to evaluate the resulting optimization model can be identified, at 206. In an embodiment, one or more KPIs represented using soft-sensors can be selected for minimizing or maximizing. Prediction models and other relevant information from the soft- sensor’s annotations are also extracted. A table of unique independent variables appearing in the prediction models of the soft-sensors may be prepared. These variables are identified either as decision variables or as disturbance variables using the knowledge base. Additionally, all those decision variables that serve the same purpose and cannot be independently manipulated are grouped together into a single category. Therefore, all the variables in a given category may be considered to be same. In case, the information is unavailable in the knowledge base, a user (for example an operator) may provide said information and is then stored in the knowledge base.

[054] At 208, decision variables, their groups (if any) and disturbance variables are identified from the unique super set of variables formulated earlier, and dynamically changing bounds on constraints, HOP and LOP, defined for certain process outputs are identified based on regime change and data (obtained from 210).

[055] Once all the variables are categorized into disturbance, decision variables and the subsequently decision variables are grouped if needed, the lower and upper bounds are extracted for the decision variables, at 212 (as will be described further with reference to FIG. 3). Herein, it will be noted that either model annotations or the knowledge base is used to gather and populate the required information. However, if the relevant information is unavailable in the knowledge base too, an operator may be notified for providing the lower and upper bounds of the decision variables manually. Along with the lower and upper bounds, the objective and constraint functions are also created at 214 from the selected soft- sensors and the other relevant information gathered at previous steps by following the process described in the FIG. 4.

[056] In general, for monitoring purposes, the prediction models are extracted from the selected soft sensors and the prediction of the key process outputs or KPIs may be performed using the actual physical sensor values and other dependent inputs to the model. However, in optimization, prediction is performed using the solver provided inputs in contrast to actual sensor values of decision variables. It should be noted that the solver provides the values of only the independent decision variables. Therefore, the values of the independent decision variables obtained from the sensors are first merged with other decision variables in each of the distinct group and finally with the current state of disturbance variables. The latter is estimated by the system using the sensor values of disturbance variables measured and communicated by the plant in real-time. Therefore, for configuring the optimization model, the prediction model is taken as an input along with information regarding the variables used in the prediction model, and the values of the decision variables. The output of the system is a computer implemented function which does the above mentioned operation and then predicts the value of a KPI for the given disturbance variables, soft- sensors selected as constraint and objective functions and inputs provided by the solver. This process is followed for both objective function and constraint functions. However, for the constraint function, the function evaluation is not the predicted value of the KPI or key process output as is the case for objective function. The output of the constraint function may be (HOP - prediction) or/and (prediction - LOP) which is a standard convention for the constraints (g(x)>0) that are accepted by the optimization solvers. Based on the above created objective functions, constraint functions (obtained from 214), and the information related to the decision variables, the optimization problem is classified in to the predefined classes, described later in text, and a corresponding solver is selected. The objective and constraint functions (obtained from 214), the optimization problem classification and the selected solver (obtained from 216), and lower and upper bounds of the decision variables (obtained from 212) are stored in the optimization model repository 218 to define the optimization model.

[057] Referring now to FIG. 3, a process flow 300 for extraction of lower and upper bounds of decision variables using the variables’ information is described in accordance with an example embodiment. In an embodiment, once all the variables are categorized into disturbance, decision variables, and the subsequently decision variables are grouped if needed (for example, at 208 of FIG. 2), the lower and upper bounds are extracted for the independent decision variables as described below.

[058] In an embodiment, the decision variables along with grouping information and disturbance variables are obtained at 302. For extracting the lower and upper bounds, at 304, it is determined whether the knowledge base has information regarding lower and upper bounds of decision variables. If it is determined at 304, that the knowledge base has information regarding lower and upper bounds of decision variables, the method 300 includes populating the lower and upper bounds of decision variables from the knowledge base at 306. If however, it is determined at 304, that the knowledge base has no information regarding lower and upper bounds of decision variables, the method 300 includes populating the lower and upper bounds of decision variables for example by user, at 308. At 310, the lower and the upper bounds of the decision variables may be stored to the optimization model (i.e. 218 of FIG. 2).

[059] As previously described, along with the lower and upper bounds, the objective and constraint functions are also created from the selected soft- sensors that are being used to optimize key process outputs or KPIs in a feasible space defined based on operational limits placed on process outputs. Each of the objective functions and the constraint functions is created using one of the plurality of soft-sensors corresponding to the KPIs or key process outputs as selected from the soft sensor repository and the accompanying annotations. In an embodiment, based on the KPIs selected by the user to configure an optimization model, a computer implemented function is created for all the selected KPIs from their corresponding soft-sensor. In this process, a prediction model is extracted from the corresponding soft-sensor of KPI or key process output as available in soft sensor repository. The extracted prediction models from corresponding soft-sensors are converted to either an objective function or a constraint function based on the input from the user. The extracted prediction models for KPIs are converted to computer implementable objective functions for every soft sensor selected as KPI with a proper sign convention. The sign convention is based on the nature of optimization of KPI that can be either maximization or minimization. One of the plurality of constraint functions are created as a computer-implemented functions from the soft sensors of the user selected KPIs, the KPIs that are to be constrained rather than optimized. The computer implemented constraint function or functions from the prediction model of the corresponding KPIs is derived in terms of standard optimization problem formulation format by accommodating at least one of the Higher Operating Point and Lower Operating Point. In case multiple KPIs are selected for optimization, the user is further prompted to decide whether to solve the problem as a multi-objective or single objective one. In case the optimization problem is to be solved as single objective optimization problem, predictions of computer implemented objective function are scaled and summed post multiplication with user defined weights to create a single objective function. In case, a plurality of soft-sensors are selected for maximization or minimization corresponding to case where multiple KPIs are to optimized, and the optimization problem is posed as multi-objective optimization problem, the computer- implemented objective functions are used as-is. In an alternate embodiment, the calculated constraint function post multiplication with predefined penalty coefficients is added to the calculated objective function when a problem comprising constraints is to be solved as unconstrained optimization problem. A process flow describing creation of objective function and constraint functions is explained further in detail with reference to FIG. 4.

[060] Referring now to FIG. 4, a flow-diagram of a method 400 for creation of objective and constraint functions using the variables’ information, available in knowledge base, and soft sensors of KPIs or key process outputs, is described in accordance with an example embodiment. In general, for monitoring purposes, the prediction models are extracted from the selected soft sensors and the prediction is performed using the actual physical sensor values and other dependent inputs to the prediction model. However, in optimization, prediction is performed using the solver provided inputs instead of using the actual sensor values of decision variables. It should be noted that the solver provides the values of only the independent decision variables. Therefore, the values of the independent decision variables from the sensors are first merged with other decision variables in each of the distinct group and finally with the current state of disturbance variables, estimated using the sensor values of disturbance variables as measured and communicated by the plant in real-time. Therefore, the information regarding the variables used in the prediction model and the values of the decision variables sent by the solver are taken as the input. The output of the method 400 is a computer implemented function which does the aforementioned operations and then predicts the output. Said process 400 is followed for both objective function and constraint functions. However, for the constraint function, the output is not as straightforward as compared to objective function. The output of the constraint function may be (HOP - prediction) or/and (prediction- LOP) which is a standard convention for the constraints (g(x)>0) that are accepted by the optimization solvers.

[061] Once the objective function and the constraint functions are created, the optimization problem type is detected by the method described in FIG. 5. In an embodiment, the user may select the type of optimization problem that has to be solved. If there are multiple objective functions, the user may solve the optimization problem as a Multi-objective optimization (MOO) problem or convert it to a single objective optimization problem. For converting a MOO to single objective optimization problem, the objective functions may be non-dimensionalized using utopia and nadir points calculated for each of the objective function. If weights are assigned to each objective function, a computer implemented function that determines the weighted sum of the objective functions is considered as an output. However, if the weights are not assigned to any objective function, single objective function is created from all the selected KPIs similar to the earlier case, but with unit weights. In case of single objective optimization, the optimization problem may be solved as an unconstraint optimization problem. In that case, the constraint functions are added to the objective function with large penalties to encourage the optimization solvers to search regions where constraints are satisfied and hence the penalty is zero. Therefore, the configuration of optimization model can produce three possible types of optimization problems: (A) Unconstrained optimization problem (B) Single objective constrained optimization problem (C) Multi objective optimization problem. The aforementioned method of detection of the type of the optimization problem is described further with reference to the method 400 below.

[062] At 402, either real time values of disturbance variables from the database or central tendency of the disturbance variable expressed in form of either mean or median combined with the dispersion tendency is either extracted or predicted as the case maybe, step 406. Herein, the real-time data may refer to data corresponding to various inputs and outputs of a plurality of manufacturing processes present in the plant. At 408, the values of disturbance and decision variables, provided by the optimization solver at 410, are merged after grouping the dependent decision variables to create an input dataset used to evaluate objective and constraint functions.

[063] At 412, the dataset and location of the independent soft sensors respectively may be passed as arguments to each selected soft sensor that is dependent in nature. The predictions models in the selected soft sensors and their dependency are extracted at 414 from the soft sensors that are to be used to formulate objective and constraint functions. Said soft sensors are retrieved at 416.

[064] At 418, it is determined whether the number of objective functions is more than one. If it is determined at 418, that the number of objective functions is more than one, then it is determined at 420, either from user or from knowledge base, whether the optimization problem is to be solved a single objective optimization. If it is determined at 420 that the optimization problem is not to be solved a single objective optimization, then at 422, the optimization problem is tagged as a multi-objective optimization problem. However, if it is determined at 420 that the optimization problem is to be solved a single objective optimization, then at 424 it is determined, either from user or from knowledge base, whether weights for the objective function are provided manually. If it is determined at 424 that the weights for the objective function are not provided manually, then at 426, a single objective function problem is created using a unit weight for each objective function. Else, a weighted objective function is created using provided weights at 428. In either case (i.e. objective function is created at 426 or 428), then method 400 follows 430. It will be noted that if at 418 the number of objective functions is determined not to be more than one, then also method 400 follows 430.

[065] At 430, it is determined whether there are any constraints based on the prior information provided by the user while defining the optimization problem. If it is determined at 430, that there are no constraints, then at 432, the optimization problem is solved as an unconstrained optimization problem. If, however, it is determined at 430, that there are constraints, then the user is asked whether the optimization problem is to be solved as an constrained optimization or an unconstrained optimization problem at 434. The latter may be required in case user wants to use an algorithm specifically written to solve unconstrained optimization problem for a constrained optimization problem. If at 434, is determined that the optimization problem is to be solved as an unconstrained optimization problem then at 438 objective function is augmented by assigning penalties to the constraints, and the objective problem is solved as a unconstrained optimization problem comprising the modified objective function. The penalties assigned are either taken as input from the user or determined by the system based on the magnitude of the objective function. [066] FIG. 5 illustrates an example flow diagram for a method 500 for classification of optimization problem, in accordance with an example embodiment. In an embodiment, a multi-level classification of the optimization problem is performed based on the variable information captured in the optimization model 502 and the knowledge base.

[067] In an embodiment, the variable information may be extracted by identifying each variable as one of disturbance variable and decision variable from the knowledge base along with the lower bound and the upper bound of the decision variable. The decision variables may be populated in groups as defined. Finally, a unified list may be created having plurality of independent decision variables and the disturbance variables for the defined optimization model.

[068] The generation/creation of the optimization model 502 is described previously with reference to FIG. 2. In an embodiment, the optimization model is passed independently to three circuits of FIG. 5. In the first circuit, the nature of objective function and constraint functions, in form of either linear or nonlinear is evaluated by sensitivity analysis. Each of the decision variables that affect the constraint and objective functions are perturbed independently and the changes in the said functions are observed. For example at 504, based on the sensitivity analysis, it is determined whether all the objective and constraint functions are linear. If at 504 it is determined that all the objective and constraint functions are linear, then the optimization problem is classified as linear programming at 508, else the optimization problem is classified as non-linear programming problem at 506.

[069] In second circuit, the optimization problem is classified either as continuous or discrete optimization based on presence of integer variables in decision variables. The discrete optimization is further classified into integer programming (IP) if all the decision variables are integer variables. Otherwise, it is classified as a mixed-integer programming (MIP) problem. For example, it is determined at 510 whether any of decision variable is an integer. If at 510 it is determined that any of the decision variables is not an integer, then at 512 the optimization problem is classified as continuous optimization problem. Else, at 514 the optimization problem is classified as discrete optimization problem. Further, at 516 it is determined whether all decision variables are integers. If it is determined at 516 that all the decision variables are integers then at 518, the optimization problem is classified as integer programming, else at 520 the optimization problem is classified as MIP.

[070] Third circuit checks for the availability of gradients either numerically or as a user input. If the gradients are available either in form of a user defined computer implemented function or are opted to be evaluated using either automatic or finite difference methods, then traditional gradient based techniques are used for optimization, else, heuristic solvers are selected for solving the above classified optimization problem. For example, at 522 availability of gradient either form user defined computer implemented function or from finite difference methods is checked. If it is determined at 522 that the gradients cannot be computed for objective and constraint functions, then at 524, heuristic solvers are selected for solving the optimization problem. If however at 522 it is determined that the gradients can be calculated for objective and constraint functions, then at 526 gradient based techniques are selected.

[071] Depending on the combination of classifications obtained using the aforementioned three circuits and information available optimization model, a solver is selected and added to the optimization model. These combinations are as mentioned below in

Table 1.

Table 1: Table with the combination of problem types

[072] A continuous and non-linear programming optimization problem is tagged as NLP problem. Similarly, a continuous and linear programming optimization problem is tagged as LP problem. Similarly, an NLP and IP are tagged as INLP problem and an NLP and MIP is tagged as MINLP problem. A LP and IP are tagged as ILP problem and a LP and MIP is tagged as MILP problem. Finally, a solver is selected depending on the above classification, optimization set-up file from the FIG. 5 and availability of the gradients. The solvers or techniques or algorithms that are used in various above mentioned classification are not limited to but as mentioned in Table 2, 3 and 4 below.

Table 2: Table with the solver/technique/method for various classifications in case of constrained single objective optimization

[073] A multi-objective NLP problem (MONLP) can be solved using nsga-2, EMOA, multi-objective particle swarm optimization techniques. In case of a multi-objective mixed integer programming (MOMILP) or a multi-objective integer programming (MOILP), heuristics like genetic algorithm based multi-objective mixed integer programming algorithm are used to solve the optimization problem. In single objective optimization, gradients are used for selecting the solver along with the type of problem defined earlier. In presence of gradients, a constrained optimization of an NLP and LP problem can be solved using techniques like Sequential Least Squares Programming (SLSQP), Quasi-Newton Optimization, etc., and unconstrained NLP and LP problem can be solved using techniques like Limited-memory BFGS-B, Trust-Region Optimization, Newton-Raphson Ridge Optimization, Quasi-Newton Optimization, and son on. However, in case the gradients are not available, both the constrained and unconstrained NLP problems can be solved using heuristic methods like real-coded genetic algorithm, particle swarm optimization, ant colony optimization, simulated annealing, etc. In case of single objective MINLP and MILP problems, techniques like branch and bound based techniques, genetic programming based techniques, etc., are selected for solving the problem.

[074] The optimization model obtained from FIGS. 2-5 (for example, the objective function/s, constraint function/s, lower and upper bounds, the solver and default/user modified solver parameters) are utilized in real-time optimization. The optimization model thus includes, but is not limited to, location of objective functions, location of constraint functions, names of decision variables, lower bounds and upper bounds of independent disturbance variables, names of dependent decision variables, names of disturbance variables, optimization solver and solver parameters.

[075] A detailed execution of real-time self-optimization is as described in FIG. 6. An optimization model is used as an input for RTO. Based on the plant condition, its requirements and various user inputs, an optimization model is configured, as explained previously. RTO is either triggered in real-time depending on any change detected during self-monitoring, or at a pre-defined frequency of time, or manually triggered by an operator.

[076] The present state of disturbance variables is required to evaluate the objective and constraint functions in optimization. However, they might not be available in the real-time to perform RTO. The reason for unavailability might be because but not limited to technical limitations of the sensors or faulty sensors or due to total time required to compute optimization being more than the time taken by the disturbance variable to change and affect the process outputs. On the other hand, the values of available disturbance variables might also have noise. Therefore, the disturbance variables information is extracted in the real-time and filtered as described with reference to FIG. 6.

[077] Referring to FIG. 6, a process flow 600 to perform real time optimization is described in accordance with the disclosed embodiments. Data preparation module sends current state of disturbance variables as an input to prepare dataset required to perform optimization using the optimization models as defined by the disclosed system. A process flow for describing extraction of disturbance variables is explained further with reference FIG. 7.

[078] The soft-sensors required for optimization in order to evaluate both objective and constraint functions are accessed from soft-sensor repository 604. The present state of disturbance variables is obtained from data preparation module. The optimization model consisting of optimization problem, optimization solver and the corresponding solver parameters is further sent as an input. The optimization model is solved/ evaluated to obtain the set points of decision variables at 602. The set-points are tested for their feasibility before being communicated to the plant at step 606. These set-points are passed to both the plant and the simulation module for execution, at 608 and 610 respectively, with an alarm to the operator in case any of the estimated set-points of the decision variable are infeasible. An operator who monitors the plant may have an access to monitor and modify the set-points. Therefore, in case the set-points are infeasible for implementation or hazardous, the operator may modify the said set-points before implementation at the plant and the simulation module. Both simulation module and the plant implement the prescribed set-points and the resultant KPIs are observed. The KPI estimated from the simulation module are called predicted KPI since they are obtained by simulation module utilizing the soft sensors, whereas the KPI coming from the plant are called realized KPIs since these are measured / observed by operating the actual plant.

[079] Referring to FIG. 7, an example process flow for a method 700 for extracting the disturbance variables for optimization is described in accordance with the various embodiments. At 702, it is determined whether disturbance variables are currently available from the plant database. In case the disturbance variables are available, the values thereof are smoothened to remove any spikes that are caused due to sensor noise using available filtering technique such as Exponential Moving Average (EMA), Extended Kalman Filter (EKF), etc. at 704. The estimated current state of the disturbance variables is compared to the previous determined state and in case the change is found to be substantial, the newly determined state of disturbance variable is populated at 706 and sent to prepare dataset for to be used for real time optimization at 712. [080] If however, the present values of the disturbance variables are not available from the plant database or the computation time for optimization is more compared to rate of change of disturbance variable, their values are estimated using time series forecasting techniques like Autoregressive integrated moving average (ARIMA), Autoregressive Integrated Moving Average with Explanatory Variable (ARMAX), Long- short- term-memory (LSTM), and so on and then passed as inputs to RTO. Hence, at 708, the central tendency and dispersion of the estimated future disturbances are predicted (for example using soft-sensors of disturbance variables obtained from soft sensor repository 710), and current or expected future setting of disturbance variable are populated at 706 to achieve approximated RTO at 712. This is referred to approximated RTO since the optimization is done with an estimated states that approximates disturbance variable over a period of time. It is applicable when solution of optimization model takes longer time compared to rate of change of disturbance variable.

[081] FIGS. 8 A and 8B illustrate flow diagrams for methods 800 and 850 respectively for enabling self-monitoring in the plant using the disclosed system, in accordance with various example embodiments. The self-monitoring system is in loop with the RTO.

[082] Referring to FIG. 8A, the performance of the actual plant and soft sensors is monitored to decide when to execute the optimization. This is accomplished by self-monitoring component/module of the system. The realized values (at 802) of KPIs from the plant and expected values (at 804) of KPIs from the simulation module obtained on implementing the decision variables prescribed by RTO system are utilized as inputs for self-monitoring in real time. The frequency of these inputs from RTO system depends on the nature of KPIs and the plant system. In general, it is decided by the engineer/operator/user. The estimated value of each KPI from the simulation module is compared to their realized KPI from the plant at 806. In order to do that, a statistical deviation metric (d) is evaluated at 808. The metric (d) may be one or more of mean, standard deviation, Euclidian distance, and so on. The selected deviation metric is monitored continuously in real time using outputs from RTO system. If the d does not exceed a pre-defined critical limit at 810, RTO is carried out at its normal pre-decided frequency at 812. However, if this deviation d exceeds the critical limit for reasons like plant regime change and so on, the user is notified to perform prediction model diagnosis so as to change/rebuilt/retune the prediction models using the soft sensor from the soft sensor repository 816.

[083] The prediction model rebuilding/retuning process can be automated or done manually offline depending on the system and the user. Finally, the updated prediction models are then saved with proper annotation as soft sensors into the soft sensor repository. The updated soft-sensors either replace the existing soft-sensors or stored separately along with the existing soft-sensors depending on whether the new soft-sensor is valid for the same operational regime as that of earlier soft- sensor. This is followed by re-configuration of the optimization model to update the changes in the decision and disturbance variables in case some variables are added or removed by model rebuilding process.

[084] Prediction model diagnosis triggered during the self-monitoring may lead to changing/rebuilding/retuning the prediction models in the soft sensor repository. When the prediction models in the soft sensors change, the optimization problem has to be reformulated as per updated prediction models. The retuning system that is proposed here takes care of the changes made in the models that are used for optimization. A detailed introduction of a method for self-optimization is explained in FIG. 9.

[085] Referring to FIG. 9, an instance of retuning operation is triggered when during monitoring, changing/rebuilding/retuning of the models in the soft sensor repository 902 occurs. When triggered, latest models related to KPI’s are identified and extracted from the soft sensor repository along with their respective annotations for soft sensors that are selected as objective and constraint functions at 904. Using these annotations, a super-set of all the variables that are required to evaluate the retuned optimization model is done at 906. The new super-set of variables is compared at 908 to the previous version of superset of variables that are earlier used in the optimization model (obtained from 910).

[086] If there is no change in the super- set of the variables based on the comparison at 908, it implies that the information regarding the features that are required for optimization is not necessary for updating the optimization model. Therefore, only the objective functions and the constraint functions of the optimization model are updated using rebuilt/retuned models/soft- sensors at 912. The updating of the objective and constraint functions are described previously with reference to FIG. 4. However, if at 908, new variables are added to the earlier superset of the variables, the relevant attributes such as variable type, data type, bounds are required to be updated. Therefore, availability of such information is checked with the knowledge base at 914. If available, this information is extracted from knowledge base and then the optimization model is updated at 916 as shown in FIG. 3. It is followed by updating the objective function and constraint functions as shown in FIG. 5. However, if the relevant information is not available in the database, the operator will be notified for configuring the optimization model at 918 as mentioned FIG. 2.

[087] Although the self-monitoring system monitors the validity of predictions using realized and expected KPIs from RTO, the validity of optimization model is monitored using a different method 850 as described in FIG. 8B. Referring to FIG. 8B, the expected KPIs (at 852) are compared (at 854) to the desired KPIs (at 856) that are stored in the knowledge base for the same plant operating regime. The expected KPIs in this case are compared in the similar way the realized and expected KPIs are compared in earlier case. For example, a statistical deviation metric (d) is evaluated at 858. The metric (d) may be mean, standard deviation, Euclidian distance, and so on. The selected deviation metric is monitored continuously in real time. If the d does not exceed a pre-defined critical limit at 860, RTO is carried out at its normal pre-decided frequency at 862. However, if this deviation d exceeds the critical limit for reasons like plant regime change and so on, the user is notified to perform optimization model diagnosis so as to change/rebuilt/retune the optimization model from the repository 864. Herein, the retuning is performed on the optimization model instead of prediction model.

[088] In case of retuning the optimization model, the solver parameters, weights used for each objective function and penalties for constraints are updated such that the expected KPIs after optimization are closer to the best possible KPIs values extracted from knowledge base for the similar scenario. A method for real-time self-optimization of manufacturing operations is described further with reference to FIGS. 10A and 10B.

[089] Referring now to FIGS. 10A and 10B, a flow diagram of a method for real-time self-optimization of manufacturing processes is described in accordance with an example embodiment. Herein, the term‘manufacturing process’ shall encompass various processes and equipment associated with a manufacturing plant. In an embodiment, the plant’s behavior may change over a period of time, and said change in behavior may be identified by monitoring disturbance variables and KPIs of the plant. Once the change in the plant exceeds a critical limit quantified in terms of changes in disturbance variables and key performance indicators, the current optimization problem is deemed no longer valid and thereby self-optimization system is triggered. In an embodiment, for self-optimization, a new optimization problem is configured by retuning either the optimization model by considering the earlier knowledge of problem formulation and the current plant conditions. The aforementioned method of real-time self-optimization is described further in detail with reference to method 1000 below.

[090] At 1002 of the method 1000, the method 1000 includes obtaining a current state of the decision variables, one or more disturbance variables and actual outputs of the plurality of manufacturing processes from a plurality of source databases. In an embodiment, the plurality of source databases comprises an optimization database, a material database, an equipment database, an operational database and a safety and environmental database. In an embodiment, in order to obtain the current state of disturbance variables, a real-time data- preprocessing of the disturbance variables is performed to identify an operating region of the disturbance variables. Based on the real-time data-preprocessing, average, standard deviation, and statistical parameters of the disturbance variable are shifted along with the operating range of the disturbance variable/variables to obtain the current state of the disturbance variables.

[091] At 1004 of the method 1000, plurality of optimal set points of decision variables for a current state of disturbance variables associated with the plurality of manufacturing processes is determined by optimizing KPIs of a plurality of manufacturing processes. The optimization of the KPIs is performed by using one or more optimization models selected from an optimization model database. An optimization model is defined by creating one or more objective function functions, one or more constraint functions (if any), based on one of the plurality of soft sensors, decision and disturbance variables along with the bounds on the decision variables, type of optimization problem, gradient information, optimization solver and appropriate solver parameters. As will be understood, the real-time self-optimization of plant operations is performed by defining optimization models for real-time optimization (RTO) so as to improve outputs of various manufacturing processes associated with the plant operation. An example of defining an optimization model is explained with reference to FIG. 2.

[092] At 1006, the method 1000 includes user selecting at least one soft-sensor from amongst a plurality of soft-sensors stored in a soft sensor repository based on the actual outputs of the plurality of manufacturing processes. As previously described, each of the plurality of soft sensors are associated with a corresponding prediction model and are capable of predicting KPI of a manufacturing process from amongst the plurality of manufacturing processes.

[093] At 1008, the method 1000 includes predicting, using the at least one soft sensor, expected output of a manufacturing process associated with the at least one soft sensor. The prediction of the expected output of the manufacturing process is described with reference to FIGS. 8 A and 8B.

[094] At 1010, the method 1000 includes comparing the actual outputs with the expected outputs to identify an offset of expected process outputs compared to actual process outputs.

[095] At 1012, the method 1000 includes comparing the expected outputs with desired outputs to identify an offset of the expected process outputs compared to the desired process outputs, the desired outputs associated with the plurality of manufacturing processes stored in a knowledge base.

[096] At 1014, the method 1000 includes enabling real-time adaptive tuning of the one or more optimization models on determination of the offset of the expected outputs compared to the desired outputs being more than a first predefined threshold. In an embodiment, real-time adaptive tuning of one of the one or more optimization models is triggered when the calculated deviation of the predicted process output from desired process output is greater than a defined threshold value. Real time adaptive tuning includes changing the solver technique for optimization, changing the hyper-parameter set of the solver technique for optimization. In an embodiment, the tuning of the optimization model includes obtaining the best hyper-parameter set of the optimization solver associated with the optimization model. For example, during the initial setup, the user might have defined some parameters for optimization solver. But in long run, as the plant condition changes, these parameters are not capable to provide optimal set points. In this scenario, the deviation metric (d) that was discussed earlier may exceed the threshold limit suggesting changing the solver parameters so that the optimal solution is achieved.

[097] At 1016, the method 1000 includes enabling real-time adaptive tuning of one or more of the plurality of soft sensors and the corresponding prediction models on determination of the offset of expected outputs compared to actual outputs being more than a second predefined threshold. In an embodiment, adaptive tuning of one of the plurality of the soft sensors includes performing at least one of changing the dataset for training the data-driven models, changing the modelling technique for training the data-driven models, changing the hyper-parameter set of the modelling technique for training the data-driven models, changing the dataset for tuning the first principle based models, tuning the model parameters for training the first principle based models, and changing the hyper -parameters of the solvers used in simulating the first principle based models.

[098] In an embodiment, for performing real-time self-optimization of the manufacturing plant operations and systems, an optimization model is configured depending on the plant and its requirements. Configuring the optimization model includes creating objective and constraint functions along with obtaining or inferring required attributes and annotation by the system 100 (FIG. 1A). The optimization model is utilized as an input to the system 100 for performing RTO. In an embodiment, the RTO is triggered in real-time depending on any change detected by self-monitoring module, or at a pre-defined frequency of time or manually triggered by an operator.

[099] FIG. 11 is a block diagram of an exemplary computer system 1101 for implementing embodiments consistent with the present disclosure. The computer system 1101 may be implemented in alone or in combination of components of the system 100 (FIG. 1). Variations of computer system 1101 may be used for implementing the devices included in this disclosure. Computer system 1101 may comprise a central processing unit (“CPU” or “hardware processor”) 1102. The hardware processor 1102 may comprise at least one data processor for executing program components for executing user- or system-generated requests. The processor may include specialized processing units such as integrated system (bus) controllers, memory management control units, floating point units, graphics processing units, digital signal processing units, etc. The processor may include a microprocessor, such as AMD Athlon™, Duron™ or Opteron™, ARM’s application, embedded or secure processors, IBM PowerPCTM, Intel’s Core, Itanium™, Xeon™, Celeron™ or other line of processors, etc. The processor 1102 may be implemented using mainframe, distributed processor, multi-core, parallel, grid, or other architectures. Some embodiments may utilize embedded technologies like application specific integrated circuits (ASICs), digital signal processors (DSPs), Field Programmable Gate Arrays (FPGAs), etc.

[0100] Processor 1102 may be disposed in communication with one or more input/output (PO) devices via I/O interface 1103. The I/O interface 1103 may employ communication protocols/methods such as, without limitation, audio, analog, digital, monoaural, RCA, stereo, IEEE- 1394, serial bus, universal serial bus (USB), infrared, PS/2, BNC, coaxial, component, composite, digital visual interface (DVI), high-definition multimedia interface (HDMI), RF antennas, S-Video, VGA, IEEE 802.11 a/b/g/n/x, Bluetooth, cellular (e.g., code-division multiple access (CDMA), high-speed packet access (HSPA+), global system for mobile communications (GSM), long-term evolution (LTE), WiMax, or the like), etc.

[0101] Using the EO interface 1103, the computer system 1101 may communicate with one or more VO devices. For example, the input device 1104 may be an antenna, keyboard, mouse, joystick, (infrared) remote control, camera, card reader, fax machine, dongle, biometric reader, microphone, touch screen, touchpad, trackball, sensor (e.g., accelerometer, light sensor, GPS, gyroscope, proximity sensor, or the like), stylus, scanner, storage device, transceiver, video device/source, visors, etc.

[0102] Output device 1105 may be a printer, fax machine, video display (e.g., cathode ray tube (CRT), liquid crystal display (LCD), light-emitting diode (LED), plasma, or the like), audio speaker, etc. In some embodiments, a transceiver 1106 may be disposed in connection with the processor 1102. The transceiver may facilitate various types of wireless transmission or reception. For example, the transceiver may include an antenna operatively connected to a transceiver chip (e.g., Texas Instruments WiLink WL1283, Broadcom BCM4750IUB8, Infineon Technologies X-Gold 618-PMB9800, or the like), providing IEEE 802.11a/b/g/n, Bluetooth, FM, global positioning system (GPS), 2G/3G HSDPA/HSUPA communications, etc.

[0103] In some embodiments, the processor 1102 may be disposed in communication with a communication network 1108 via a network interface 1107. The network interface 1107 may communicate with the communication network 1108. The network interface may employ connection protocols including, without limitation, direct connect, Ethernet (e.g., twisted pair 10/100/1000 Base T), transmission control protocol/internet protocol (TCP/IP), token ring, IEEE 802.1 la/b/g/n/x, etc. The communication network 1108 may include, without limitation, a direct interconnection, local area network (LAN), wide area network (WAN), wireless network (e.g., using Wireless Application Protocol), the Internet, etc. Using the network interface 1107 and the communication network 408, the computer system 1101 may communicate with devices 1109 and 1110. These devices may include, without limitation, personal computer(s), server(s), fax machines, printers, scanners, various mobile devices such as cellular telephones, smartphones (e.g., Apple iPhone, Blackberry, Android-based phones, etc.), tablet computers, eBook readers (Amazon Kindle, Nook, etc.), laptop computers, notebooks, gaming consoles (Microsoft Xbox, Nintendo DS, Sony PlayStation, etc.), or the like. In some embodiments, the computer system 401 may itself embody one or more of these devices.

[0104] In some embodiments, the processor 1102 may be disposed in communication with one or more memory devices (e.g., RAM 713, ROM 714, etc.) via a storage interface 1112. The storage interface may connect to memory devices including, without limitation, memory drives, removable disc drives, etc., employing connection protocols such as serial advanced technology attachment (SATA), integrated drive electronics (IDE), IEEE-1394, universal serial bus (USB), fiber channel, small computer systems interface (SCSI), etc. The memory drives may further include a drum, magnetic disc drive, magneto-optical drive, optical drive, redundant array of independent discs (RAID), solid-state memory devices, solid-state drives, etc. Variations of memory devices may be used for implementing, for example, any databases utilized in this disclosure.

[0105] The memory devices may store a collection of program or database components, including, without limitation, an operating system 1116, user interface application 1117, user/application data 1118 (e.g., any data variables or data records discussed in this disclosure), etc. The operating system 1116 may facilitate resource management and operation of the computer system 1101. Examples of operating systems include, without limitation, Apple Macintosh OS X, Unix, Unix-like system distributions (e.g., Berkeley Software Distribution (BSD), FreeBSD, NetBSD, OpenBSD, etc.), Linux distributions (e.g., Red Hat, Ubuntu, Kubuntu, etc.), IBM OS/2, Microsoft Windows (XP, Vista/7/8, etc.), Apple iOS, Google Android, Blackberry OS, or the like. User interface 1117 may facilitate display, execution, interaction, manipulation, or operation of program components through textual or graphical facilities. For example, user interfaces may provide computer interaction interface elements on a display system operatively connected to the computer system 1101, such as cursors, icons, check boxes, menus, scrollers, windows, widgets, etc. Graphical user interfaces (GUIs) may be employed, including, without limitation, Apple Macintosh operating systems’ Aqua, IBM OS/2, Microsoft Windows (e.g., Aero, Metro, etc.), Unix X-Windows, web interface libraries (e.g., ActiveX, Java, Javascript, AJAX, HTML, Adobe Flash, etc.), or the like.

[0106] In some embodiments, computer system 1101 may store user/application data 1118, such as the data, variables, records, etc. as described in this disclosure. Such databases may be implemented as fault-tolerant, relational, scalable, secure databases such as Oracle or Sybase. Alternatively, such databases may be implemented using standardized data structures, such as an array, hash, linked list, structured text file (e.g., XML), table, or as hand-oriented databases (e.g., using HandStore, Poet, Zope, etc.). Such databases may be consolidated or distributed, sometimes among the various computer systems discussed above in this disclosure. It is to be understood that the structure and operation of any computer or database component may be combined, consolidated, or distributed in any working combination.

[0107] Additionally, in some embodiments, the server, messaging and instructions transmitted or received may emanate from hardware, including operating system, and program code (i.e., application code) residing in a cloud implementation. Further, it should be noted that one or more of the systems and methods provided herein may be suitable for cloud-based implementation. For example, in some embodiments, some or all of the data used in the disclosed methods may be sourced from or stored on any cloud computing platform.

[0108] Various embodiments disclosed herein provided method and system for real time self-optimization of manufacturing operations and systems. Conventional system and methods for plant optimization lack the ability to perform plant-wide optimization using predictive models. Furthermore, a system for real time optimization and its self-adaptation too is not available. The disclosed embodiments proposes method and system optimizes the key process outputs using predictive models and information already available in plant database. Furthermore, critical events required for self-tuning of optimization problem to suit the current plant and environmental state are identified to facilitate real-time self-optimization. In an embodiment, the disclosed system can self-identify critical events such as change in the behavior of disturbance variables, change in the models that govern the objective and constraints, and so on. In response, the disclosed system automatically creates objective and constraint functions from predictive models and supporting information. The disclosed system can identify and classify the type of optimization problem that needs to be solved such as single objective, multi-objective, integer programming, mixed integer programming, and identifies an appropriate optimization solver and solver parameters.

[0109] The written description describes the subject matter herein to enable any person skilled in the art to make and use the embodiments. The scope of the subject matter embodiments is defined by the claims and may include other modifications that occur to those skilled in the art. Such other modifications are intended to be within the scope of the claims if they have similar elements that do not differ from the literal language of the claims or if they include equivalent elements with insubstantial differences from the literal language of the claims.

[0110] It is to be understood that the scope of the protection is extended to such a program and in addition to a computer-readable means having a message therein; such computer-readable storage means contain program-code means for implementation of one or more steps of the method, when the program runs on a server or mobile device or any suitable programmable device. The hardware device can be any kind of device which can be programmed including e.g. any kind of computer like a server or a personal computer, or the like, or any combination thereof. The device may also include means which could be e.g. hardware means like e.g. an application- specific integrated circuit (ASIC), a field- programmable gate array (FPGA), or a combination of hardware and software means, e.g. an ASIC and an FPGA, or at least one microprocessor and at least one memory with software processing components located therein. Thus, the means can include both hardware means and software means. The method embodiments described herein could be implemented in hardware and software. The device may also include software means. Alternatively, the embodiments may be implemented on different hardware devices, e.g. using a plurality of CPUs.

[0111] The embodiments herein can comprise hardware and software elements. The embodiments that are implemented in software include but are not limited to, firmware, resident software, microcode, etc. The functions performed by various components described herein may be implemented in other components or combinations of other components. For the purposes of this description, a computer-usable or computer readable medium can be any apparatus that can comprise, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. [0112] The illustrated steps are set out to explain the exemplary embodiments shown, and it should be anticipated that ongoing technological development will change the manner in which particular functions are performed. These examples are presented herein for purposes of illustration, and not limitation. Further, the boundaries of the functional building blocks have been arbitrarily defined herein for the convenience of the description. Alternative boundaries can be defined so long as the specified functions and relationships thereof are appropriately performed. Alternatives (including equivalents, extensions, variations, deviations, etc., of those described herein) will be apparent to persons skilled in the relevant art(s) based on the teachings contained herein. Such alternatives fall within the scope of the disclosed embodiments. Also, the words “comprising,” “having,” “containing,” and “including,” and other similar forms are intended to be equivalent in meaning and be open ended in that an item or items following any one of these words is not meant to be an exhaustive listing of such item or items, or meant to be limited to only the listed item or items. It must also be noted that as used herein and in the appended claims, the singular forms“a,”“an,” and“the” include plural references unless the context clearly dictates otherwise.

[0113] Furthermore, one or more computer-readable storage media may be utilized in implementing embodiments consistent with the present disclosure. A computer-readable storage medium refers to any type of physical memory on which information or data readable by a processor may be stored. Thus, a computer-readable storage medium may store instructions for execution by one or more processors, including instructions for causing the processor(s) to perform steps or stages consistent with the embodiments described herein. The term“computer-readable medium” should be understood to include tangible items and exclude carrier waves and transient signals, i.e., be non-transitory. Examples include random access memory (RAM), read-only memory (ROM), volatile memory, nonvolatile memory, hard drives, CD ROMs, DVDs, flash drives, disks, and any other known physical storage media.

[0114] It is intended that the disclosure and examples be considered as exemplary only, with a tme scope of disclosed embodiments being indicated by the following claims.

Claims

1. A processor implemented method for real-time self-optimization of manufacturing processes in a manufacturing plant, the method comprising:

obtaining, via the one or more hardware processors, a current state of one or more decision variables, one or more disturbance variables, and actual process outputs of a plurality of manufacturing processes at the manufacturing plant from a plurality of source databases and one or more soft sensors from amongst a plurality of soft sensors;

determining, via the one or more hardware processors, a plurality of optimal set points of the one or more decision variables for the current state of the one or more disturbance variables associated with the plurality of manufacturing processes for optimization of outputs of the plurality of manufacturing processes, the optimization performed by using one or more optimization models selected from an optimization model database;

selecting, via the one or more hardware processors, at least one soft sensor from amongst the plurality of soft sensors stored in a soft sensor repository corresponding to the actual process outputs of the plurality of manufacturing processes, each of the plurality of soft sensors associated with a corresponding prediction model and capable of predicting process output of a manufacturing process from amongst the plurality of manufacturing processes; predicting, using the at least one soft sensor corresponding to the actual process outputs that are being optimized, expected process output of a manufacturing process associated with the at least one soft sensor, via the one or more hardware processors;

comparing the actual process outputs with the expected process outputs to calculate a deviation of the expected process outputs from the actual process outputs, via the one or more hardware processors;

comparing, via the one or more hardware processors, the expected process outputs with desired process outputs to calculate deviation of the expected process outputs from the desired process outputs, the desired process outputs associated with the plurality of process outputs in manufacturing processes stored in a knowledge base;

enabling, via the one or more hardware processors, real-time adaptive tuning of the one or more optimization models on determination of the deviation of the expected process outputs compared to the desired process outputs being more than a first predefined threshold; and

enabling, via the one or more hardware processors, one of a generation of a request to replace the soft sensor and real-time adaptive tuning of one or more of the plurality of soft sensors and the corresponding prediction models on determination of the deviation of expected process outputs compared to actual process outputs being more than a second predefined threshold.

2. The method of claim 1, wherein obtaining the current state of the disturbance variables comprises:

identifying an operating region of the disturbance variables based on a real-time data- preprocessing of the disturbance variables; and

shifting, based on the real-time data-preprocessing, average, standard deviation, and statistical parameters of the disturbance variable along with an operating region of the disturbance variable to obtain the current state of the disturbance variables.

3. The method of claim 1, wherein a soft sensor of a process output in the soft sensor repository comprises annotations needed by the one or more optimization models, and a prediction model extracted from a prediction model repository, wherein the prediction model extracted from the prediction model repository comprises at least one of physics based model, data-driven model, and hybrid model, wherein the hybrid model is a combination of physics based model and data-driven model, and wherein one of the plurality of prediction models extracted from the prediction model repository predicts one of the plurality of the process outputs.

4. The method of claim 3, wherein the annotations corresponding to one of the plurality of soft-sensors comprise name of process output, prediction model to predict the process output, one of the plurality of names of variables, and model parameters that are used to build the corresponding prediction model, one or more soft-sensors amongst a plurality of the independent soft sensors used to build the soft sensor, and data used to build the prediction model.

5. The method of claim 1, further comprising defining an optimization model from amongst the one or more optimization models for real-time self-optimization to improve the process outputs, wherein defining the optimization model comprises creating one or more objective functions and one or more constraint functions based on one or more soft sensors amongst the plurality of soft sensors, defining decision variables and disturbance variables along with lower bound and upper bound on the decision variables, identifying type of optimization problem, estimating gradient information, and identifying optimization solver and solver parameters.

6. The method of claim 5, wherein each of the one or more objective functions and the one or more constraint functions are created using variables’ information from a knowledge base associated with the and one or more soft sensors from amongst the plurality of soft- sensors selected from the soft sensor repository, and wherein creating an objective function of the one or more objective functions and a constraint function of the one or more constraint functions comprises:

extracting one or more prediction models and the associated annotations comprising information of process variables from the one or more soft-sensors of process outputs; converting the extracted one or more prediction models from the corresponding one or more soft-sensors to the objective function and the constraint functions by:

creating at least one objective function from the corresponding one or more soft sensors depending on whether each of the corresponding process output is targeted to be one of maximized and minimized;

creating at least one constraint function from the corresponding one or more soft sensors and pre-defined higher operating point and lower operating point of the corresponding process outputs, in case one or more process outputs are to be constrained;

scaling the predictions of the one or more soft sensors corresponding to the process outputs to be optimized and summing up the scaled predictions with pre-defined weights to evaluate objective function when one or more of process outputs are to be optimized and optimization problem is to be solved as single objective optimization problem;

using each of the predictions of the one or more soft-sensors corresponding to the key process outputs as a separate evaluation of corresponding objective function when plurality of process outputs are to be optimized and optimization problem is to be solved as a multi-objective optimization problem; and

adding, to the evaluated objective function, the evaluated constraint functions post multiplication with predefined penalty coefficients, when optimization problem is to be solved as a unconstrained optimization problem.

7. The method of claim 5, further comprising extracting variable information, wherein extracting the variable information comprises: identifying each variable as one of disturbance variable and decision variable from the knowledge base along with the lower bound and the upper bound of the decision variable; populating the decision variables in groups as defined; and

creating a unified list comprising a plurality of independent decision variables and the disturbance variables for the defined optimization model.

8. The method of claim 7, further comprising performing multi-level classification of the optimization problem based on the variable information captured in the optimization model and the knowledge base, wherein performing the multi-level classification comprises: performing first level classification of the optimization problem associated with the real-time self-optimization as one of single objective unconstrained, multi objective, or single objective constrained optimization problem;

performing second level classification of the optimization problem associated with the real-time self-optimization as one of Non Linear Programming (NLP) and a Linear Programming (LP); and

performing third level classification of the optimization problem associated with the real-time self-optimization as one of an integer programming (IP) problem and a mixed- integer programming (MIP) problem.

9. The method of claim 5, further comprising identifying whether each of the one or more the objective functions and the one or more constraint functions is one of a continuous function and a discrete function.

10. The method of claim 9, wherein gradient information comprises, one of storing exact gradients, and numerically estimating the gradients when the one or more objective function and the one or more constraint function are determined to be continuous.

11. The method of claim 1, wherein the knowledge base comprises a plurality of soft sensors and corresponding annotations, datasets used to build optimization models, names of available variables and parameters, minimum and maximum of variables and parameters for different regimes of data, variable type (one of decision variable and disturbance variable), data type of each variable (one of discrete data type and continuous data type), desired process outputs for different regimes of operation of the manufacturing plant, and the corresponding optimal set-points of decision variables.

12. The method of claim 1, wherein real-time adaptive tuning of the one or more optimization models is triggered when the calculated deviation of the predicted process output from the desired process output is greater than a defined threshold value.

13. The method of claim 12, wherein the real-time adaptive tuning of the optimization model comprises obtaining at least one of a parameter set and hyper-parameter set of an optimization solver associated with the optimization model.

14. The method of claim 1, wherein the real-time adaptive tuning of one or plurality of the soft sensors comprises one of tuning the model parameters of the corresponding prediction models based on plant data when the prediction model is mechanistic or phenomenological in nature, tuning the hyper-parameters of the solvers used in simulating the mechanistic or phenomenological prediction models, altering of one or more of dataset for training data- driven models, tuning hyper-parameter set of the modelling technique for training the data- driven models, or altering modelling technique for training the data-driven models.

15. A system (1100) for real-time self-optimization of manufacturing processes in a manufacturing plant, comprising:

one or more memories (1115); and

one or more hardware processors (1102), the one or more memories (1115) coupled to the one or more hardware processors (1102), wherein the one or more hardware processors (1102) are capable of executing programmed instructions stored in the one or more memories (1115) to:

obtain a current state of the one or more decision variables and one or more disturbance variables, and actual process outputs of the plurality of manufacturing processes from a plurality of source databases;

determine a plurality of optimal set points of the one or more decision variables for a current state of the one or more disturbance variables associated with the plurality of manufacturing processes for optimization of outputs of the plurality of manufacturing processes, the optimization performed by using one or more optimization models selected from an optimization model database;

select at least one soft sensor from amongst a plurality of soft sensors stored in a soft sensor repository based on the actual process outputs of the plurality of manufacturing processes, each of the plurality of soft sensors associated with a corresponding prediction model and capable of predicting output of a manufacturing process from amongst the plurality of manufacturing processes;

predict, using the at least one soft sensor corresponding to the actual process outputs that are being optimized, expected process output of a manufacturing process associated with the at least one soft sensor;

compare the actual process outputs with the expected process outputs to identify a deviation of the expected process outputs compared to the actual process outputs;

compare the expected process outputs with desired process outputs to identify a deviation of the expected process outputs compared to the desired process outputs, the desired process outputs associated with the plurality of manufacturing processes stored in a knowledge base;

enable real-time adaptive tuning of the one or more optimization models on determination of the deviation of the expected process outputs compared to the desired process outputs being more than a first predefined threshold; and

enable one of a generation of a request to replace the soft sensor and real-time adaptive tuning of one or more of the plurality of soft sensors and the corresponding prediction models on determination of the deviation of expected process outputs compared to actual process outputs being more than a second predefined threshold.

16. The system of claim 15, wherein to obtain the current state of the disturbance variables, the one or more hardware processors are configured by the instructions to:

identify an operating region of the disturbance variables based on a real-time data- preprocessing of the disturbance variables; and

shift, based on the real-time data-preprocessing, average, standard deviation, and statistical parameters of the disturbance variable along with an operating region of the disturbance variable to obtain the current state of the disturbance variables.

17. The system of claim 15, wherein a soft sensor of a process output in the soft sensor repository comprises annotations needed by the one or more optimization models, and a prediction model extracted from a prediction model repository, wherein the prediction model extracted from the prediction model repository comprises at least one of physics based model, data-driven model, and hybrid model, wherein the hybrid model is a combination of physics based model and data-driven model, and wherein one of the plurality of prediction models extracted from the prediction model repository predicts one of the plurality of the process outputs.

18. The system of claim 17, wherein the annotations corresponding to one of the plurality of soft-sensors comprise name of process output, prediction model to predict the process output, one of the plurality of names of variables, and model parameters that are used to build the corresponding prediction model, one or more soft-sensors amongst a plurality of the independent soft sensors used to build the soft sensor, and data used to build the prediction model.

19. The system of claim 15, the one or more hardware processors are further configured by the instructions to define an optimization model from amongst the one or more optimization models for real-time self-optimization to improve the process outputs, wherein to define the optimization model, the one or more hardware processors are configured by the instructions to create one or more objective functions and one or more constraint functions based on one or more soft sensors amongst the plurality of soft sensors, define decision variables and disturbance variables along with lower bound and upper bound on the decision variables, identify type of optimization problem, estimating gradient information, and identify optimization solver and solver parameters.

20. The system of claim 19, wherein each of the one or more objective functions and the one or more constraint functions are created using variables’ information from a knowledge base associated with the and one or more soft sensors from amongst the plurality of soft- sensors selected from the soft sensor repository, and wherein to create an objective function of the one or more objective functions and a constraint function of the one or more constraint functions, the one or more hardware processors are configured by the instructions to: extract one or more prediction models and the associated annotations comprising information of process variables from the one or more soft-sensors of process outputs; convert the extracted one or more prediction models from the corresponding one or more soft-sensors to the objective function and the constraint functions by:

create at least one objective function from the corresponding one or more soft sensors depending on whether each of the corresponding process output is targeted to be one of maximized and minimized; create at least one constraint function from the corresponding one or more soft sensors and pre-defined higher operating point and lower operating point of the corresponding process outputs, in case one or more process outputs are to be constrained;

scale the predictions of the one or more soft sensors corresponding to the process outputs to be optimized and summing up the scaled predictions with pre-defined weights to evaluate objective function when one or more of process outputs are to be optimized and optimization problem is to be solved as single objective optimization problem;

use each of the predictions of the one or more soft-sensors corresponding to the key process outputs as a separate evaluation of corresponding objective function when plurality of process outputs are to be optimized and optimization problem is to be solved as a multi-objective optimization problem; and

add, to the evaluated objective function, the evaluated constraint functions post multiplication with predefined penalty coefficients, when optimization problem is to be solved as a unconstrained optimization problem.

21. The system of claim 19, wherein the one or more hardware processors are further configured by the instructions to extract variable information, wherein extracting the variable information comprises:

identify each variable as one of disturbance variable and decision variable from the knowledge base along with the lower bound and the upper bound of the decision variable; populate the decision variables in groups as defined; and

create a unified list comprising a plurality of independent decision variables and the disturbance variables for the defined optimization model.

22. The system of claim 21, wherein the one or more hardware processors are further configured by the instructions to perform multi-level classification of the optimization problem based on the variable information captured in the optimization model and the knowledge base, wherein to perform the multi-level classification, the one or more hardware processors are further configured by the instructions to:

perform first level classification of the optimization problem associated with the real time self-optimization as one of single objective unconstrained, multi objective, or single objective constrained optimization problem; perform second level classification of the optimization problem associated with the real-time self-optimization as one of Non Linear Programming (NLP) and a Linear Programming (LP); and

perform third level classification of the optimization problem associated with the real time self-optimization as one of an integer programming (IP) problem and a mixed-integer programming (MIP) problem.

23. The system of claim 19, wherein the one or more hardware processors are further configured by the instructions to identify whether each of the one or more the objective functions and the one or more constraint functions is one of a continuous function and a discrete function.

24. The system of claim 23, wherein gradient information comprises, one of storing exact gradients, and numerically estimating the gradients when the one or more objective function and the one or more constraint function are determined to be continuous.

25. The system of claim 16, wherein the knowledge base comprises a plurality of soft sensors and corresponding annotations, datasets used to build optimization models, names of available variables and parameters, minimum and maximum of variables and parameters for different regimes of data, variable type (one of decision variable and disturbance variable), data type of each variable (one of discrete data type and continuous data type), desired process outputs for different regimes of operation of the manufacturing plant, and the corresponding optimal set-points of decision variables.

26. The system of claim 16, wherein real-time adaptive tuning of the one or more optimization models is triggered when the calculated deviation of the predicted process output from the desired process output is greater than a defined threshold value.

27. The system of claim 26, wherein the real-time adaptive tuning of the optimization model comprises obtaining at least one of a parameter set and hyper-parameter set of an optimization solver associated with the optimization model.

28. The system of claim 16, wherein to perform the real-time adaptive tuning of one or plurality of the soft sensors, the one or more hardware processors are further configured by the instructions to perform one of tune the model parameters of the corresponding prediction models based on plant data when the prediction model is mechanistic or phenomenological in nature, tune the hyper-parameters of the solvers used in simulating the mechanistic or phenomenological prediction models, alter of one or more of dataset for training data-driven models, tune hyper-parameter set of the modelling technique for training the data-driven models, or alter modelling technique for training the data-driven models.

29. A computer program product comprising a non-transitory computer readable medium having a computer readable program embodied therein, wherein the computer readable program, when executed on a computing device, causes the computing device to:

obtain, via the one or more hardware processors, a current state of one or more decision variables, one or more disturbance variables, and actual process outputs of a plurality of manufacturing processes at the manufacturing plant from a plurality of source databases and one or more soft sensors from amongst a plurality of soft sensors;

determine, via the one or more hardware processors, a plurality of optimal set points of the one or more decision variables for the current state of the one or more disturbance variables associated with the plurality of manufacturing processes for optimization of outputs of the plurality of manufacturing processes, the optimization performed by using one or more optimization models selected from an optimization model database;

select, via the one or more hardware processors, at least one soft sensor from amongst the plurality of soft sensors stored in a soft sensor repository corresponding to the actual process outputs of the plurality of manufacturing processes, each of the plurality of soft sensors associated with a corresponding prediction model and capable of predicting process output of a manufacturing process from amongst the plurality of manufacturing processes;

predict, using the at least one soft sensor corresponding to the actual process outputs that are being optimized, expected process output of a manufacturing process associated with the at least one soft sensor, via the one or more hardware processors;

compare the actual process outputs with the expected process outputs to calculate a deviation of the expected process outputs from the actual process outputs, via the one or more hardware processors;

compare, via the one or more hardware processors, the expected process outputs with desired process outputs to calculate deviation of the expected process outputs from the desired process outputs, the desired process outputs associated with the plurality of process outputs in manufacturing processes stored in a knowledge base;

enable, via the one or more hardware processors, real-time adaptive tuning of the one or more optimization models on determination of the deviation of the expected process outputs compared to the desired process outputs being more than a first predefined threshold; and

enable, via the one or more hardware processors, one of a generation of a request to replace the soft sensor and real-time adaptive tuning of one or more of the plurality of soft sensors and the corresponding prediction models on determination of the deviation of expected process outputs compared to actual process outputs being more than a second predefined threshold.